深部地下结构随机声场与结构损伤智能识别

深部地下结构是国家深地战略的重要支撑。深部地下的高应力作用，导致结构损伤加剧，开展其损伤的智能识别有重要价值。目前，超声波的探测精度始终距离工程领域的高精度要求有较大差距。研究表明，结构损伤体现为混凝土中的裂隙聚集区。通过超声检测方法，识别定位混凝土中的单一微裂隙的是困难的，但是通过超声信号（如衰减参数等）识别结构损伤所表现的裂隙聚集区是可行的。本课题提出开展真三轴应力条件下的超声参数和超声尾波测试；在此基础上，建立混凝土和结构损伤随机介质模型，通过随机声场的蒙特卡洛数值模拟，研究混凝土及其损伤的超声波场的变化规律，提取其反射波和超声尾波信号，建立海量的超声参数和结构损伤的数据样本库，并开展试验测试；采用数据样本对深度学习的神经网络进行训练，逐次迭代、不断验证，实现结构损伤的智能识别。本项目研究可望突破结构超声检测的物性基础和数据处理瓶颈，对工程结构、尤其是深部地下结构智能监测有重要意义。

It is of great value to carry out intelligent damage identification for deep underground structure, because of the irregular engineering disasters. At present, the ultrasonic detection accuracy is always far from the high precision requirements in engineering field. Research shows that the damage of concrete structure is manifested as the crack accumulation area. It is difficult to characterize and locate a single micro-crack in concrete by ultrasonic detection method, but it is feasible to identify the crack accumulation area caused by structural damage through ultrasonic signals (ie. attenuation parameters). This program proposed to carry out ultrasonic parameters and ultrasonic coda wave testing under the condition of true triaxial stress. On this basis, the random medium and medium damage model of concrete were established. The evolution features of concrete damage and its ultrasonic field were studied through Monte Carlo numerical simulation of sound wave field. In addition, the ultrasonic coda of structural damage was extracted to establish a massive sample database containing ultrasonic parameters and coda wave records. Then, the deep learning method of convolutional neural network is adopted to improve the accuracy of intelligent structural damage identification through training and verification. The study of this project greatly improves the efficiency and accuracy of NDT for the intelligent monitoring of in-service concrete structures, especially for the deep underground structures.